A little status report;

Well I am starting to get a good feel for the accuracy and repeatability of the equipment. In terms of accuracy of the sensors and the calibrations for them I am pretty happy, I am satisfied that they are accurate where they need to be and that they are being consistant.
One exception is the flowrate calculation at higher flows, it uses the water in-out delta T and places impossible demands on this number above about 5-6lpm. I would need a resolution in excess of 0.001° in order to generate an accurate number at these rates, that at a heat input of 60W. At lower power levels the problem gets even worse.
Still, this gives me info about the water temperature accuracy, by varying input power I can detect these small errors and make a correction to at least keep the flowrate the same over varying power levels. This is a really small correction 1 or 2 lsb's (0.005-0.01°C) and has absolutely no visible impact on for example waterblock C/W. This particular calculation (flowrate) is also affected by the calculated heat flux but the sensitivity is far less, requiring a die sensor offset in excess of 0.4°C to produce the same error. I choose to believe that I am far better than this, accuracy for these low resolution sensors being better than I am able to resolve (~0.025°C for "in copper" thermistors, ~0.005°C for water sensors) That might sound a little weird, but with averaging you can exceed the resolution of the system in accuracy, if the measured quantity varies in absolute terms. (Perturbation theory anyone?) Also the Steinhart-Hart system for calibrating thermistors is industry standard and has measured errors in the region of less than 0.001°.


What I am not happy with is the mechanics of my setup. I am finding it extremely difficult to mount properly with the smaller die, I am getting tilted mounts rather frequently, this will require lots of practice, remounts and new thermal paste before I get confident of repeatablity. The previous die at 12x12mm was an order of magnitude easier despite not having the fluxblock alignment supports I have now. I could get the same results every time with little effort. This I would put foward as the single biggest reason for NOT using a 10x10mm die for testing. However, for now I guess I just need practice.

Warning about the above series of posts, these have all been for the purpose of verifying the equipment and are the results of ONE mount. I am not confident that this was a "good" mount so expect future readings to vary a bit. I will built up a mount history before posting any results as absolute.

I am confident that it is possible to generate absolute numbers using this technique.

I seem to be getting the hang of it. Best system seems to be to "dock" the die to the block keeping it flat by eye as the screws are tightened then turn on the heat. By minimising the TIM C/W readout the numbers seem to fall into place. Several days of this to go yet though, I am also concerned that I am cranking down a little too hard but it seems to be the only way to prevent WB C/W number giving nonsense. TIM measurements are going to be difficult.

Will leave you to the joys of testing
I Will retire to the pub for a couple of days.
Then ponder my niggles(downgraded from grave)
I think they are centred on "why holes lead to 5 -> 8.5%(^ with LPM) error in C/W(wb) and a constant 1.2 % error in C/W)TIM)" May be obvious to you but it eludes me.
You can move progress in peace for at least a week


Edit:
Changed
"why one hole ~2mm from wb/bloc IF leads to 5 -> 8.5%(^ with LPM) error and two holes ~2mm from bloc/die IF lead to a constant 1.2 % error"
to
"why holes lead to 5 -> 8.5%(^ with LPM) error in C/W(wb) and a constant 1.2 % error in C/W)TIM)"

Edit
Deleted incorrect graph which was leading nowhere

Interesting observation Les. I guess it should be constant? Might be indicative of an error somewhere, I'll work this when I get time.

Off to pub.
Added one last plot.

Edit1
Miscalculated x-axis (used 10000/C/W(wb) should be 10000/C/W(wb+TIM))
Corrected.
Now going pub.

Edit2
Deleted : should be using "Mean Twater" for both h(TIM/fb) and h(eff) and for h(eff) should be using L"(I think)

Stupid me, should be considering virtual distances for water
Or at least some "mean temp" water

The niggles amplify

Edit
Have deleted offending graphs
Niggles subsiding as I develop/digest attached model

Giving chasing shadowy figures a break.

Maybe should include dP*Q correction for calculating flow rate from Heat Flux
Only ~ 1.5w here but......
Heat Dump =HeatFlux + KPa*LPM/60
LPM=HeatFlux/(69.77*dT - Kpa/60)......(Cp=4186 J/KgC) Think 60 is rigorous but not checked)

Dunno what it does to your queried "dp correction?= -0.05"

Edit: As usual, corrected graph.

This I didn't think of Les. I have not explored this at all, is it that how it works, KPa*lpm/60 = extra heat?

Have not been here. I will check this out.
There is definitely an offset on the dP readings. I check by decreasing flow to zero and the reading goes negative. This is a constantly changing value, I need to fix some leaks, it could also be a waterblock dependent offset, certainly a "fill of the system" dependent parameter. Easy to check before each measurement though. If it's affecting flow rate readings I need to pay more attention to it.

Yes
First raised by Tecumseh and discussed here
Was thinking more as something to be confirmed when you are rich and get a flow meter.

Where I am at after 5 days pondering.

I can probably extract the average temperature of a "cut" through the hole centre from the Femlab model. Would that be more appropriate do you think?

Yes

More:
Accept modeling of fluxbloc-die is required to account for holes

Think centreline Femlab is best option to identify with measured Ts .
With proviso that h(eff) fb-surface effect on Ls should be taken into account.Range which interests us here is ~45,000 to 70,000*

Which configuration is the L set "0.000, 1.976, 3.953, 3.953,6.030,11.105,13.456" obtained? "h(eff)=20,000 on 50x50x5 bp" gives h(eff) fb-surface=77,519W/m^2C, whilst "h(eff)=50,000 on 30x30x2bp" gives h(eff) fb-surface=88,496. If both give same set may have issue with their modeling of bp, terminology, or suspect accuracy of model.

Think centreline Femlab is inappropriate for C/W(wb+TIM)
Here an average fb-surface is required.

* Edit1 - forgot TIM Edit2- forgot water temp rise (sig at low h) Edit3 -None yet

Managed to get some numbers out. The model is quite dense so I won't post all the data (over 60000 points)
Note that the distribution of the element vertices is not uniform (see attached example, T die low), hence perhaps median is better? Anyway:
L_____ 00.007 00.000 00.002 -00.002 -00.004
Mean__ 52.076 40.089 43.562 36.689 32.816
Median 52.063 40.090 43.566 36.714 32.953
Stdev_ 00.094 00.035 00.018 00.135 00.633
Range_ 00.396 00.143 00.080 00.538 03.234
Min___ 51.947 40.004 43.513 36.329 30.342
Max___ 52.343 40.147 43.593 36.867 33.576

Haven't got time to do anything with this right now, be my guest Les.

Only just decided on the question.
The one I will try to answer :-
Knowing Ts 1mm up from centreline at positions -2, 2 and 7; use Femlab simulation to determine Tmedian,Tmean,Tmin,and Tmax at position -4.23.?

Right question.........?
Ugh

Is probe physically 1mm up fom centreline or centreline?
Even if centreline will T be more representative of "1mm up" for Femlab interpretation ?
Ugh
Going to the pub, back in a couple of days

Dunno, but think it depends on the outcome of the "basis of "C" in C/W ?" thread[,when it finally gets round to wbs.
With it appearing that most die are going to have holes use whichever C is decided for calculating (C/W)in.
For Rwb am in favour of using(link)
Q=UAdT(MTD)

and equating Rwb to UA (Rwb=1/UA)
Where
A=wb/die interface area
U="Overall Heat Transfer Coefficient"
and dT(MTD)= possibly LMTD= (Two -Twi)/lin((Tdie - Twi)/(Tdie-Two))
where Tdie is either mean or median,
So again, Dunno.

Getting nowhere with modeling "fb's surface T".
Think need Femlab surface solution
Alternatively an adaptation of your Model, getting Femlab solution for TIM and using in the "adapted Model" to obtain it's W, then using to get a Tfb.surface.
Keep returning to your old Model ,thinking this is the answer but using columns instead of slices, muttering incantations and waiting.Nothing happens, think it needs your magic.

Edit: Inserted Rwb=1/UA